Abstract
Despite the significant benefits that the 6G‐enabled massive Internet of Things (IoT) applications will bring to the economy and society in the coming years, it is expected that the exponential increase in the number and the diversity of IoT devices and connections in the massive IoT ecosystem will raise a plethora of known and unknown security and privacy challenges for these applications in the 6G era. Consequently, novel security solutions effectively protecting massive IoT applications from future adversaries, while taking into consideration the resource‐constrained characteristics of the massive IoT ecosystem and the stringent network performance and privacy‐preserving requirements of these applications, are critical for their acceptance and wide adoption in the upcoming 6G era. In this context, Federated Learning (FL)‐based intrusion detection is a promising solution for effective intrusion detection in the massive IoT ecosystem, as it scales well with the massive growth of resource‐constrained IoT devices and the wide geographical spread of generated IoT data across wide‐area IoT networks. In addition, the FL approach enables local model training at each client based on its locally available training dataset instead of sending it to a remote central server (i.e. centralized intrusion detection), which may bring single‐point failure risks and compromise the privacy of the dataset. Furthermore, the aggregation of locally trained models, supported by FL, allows the quick development of accurate models even for devices generating only few training data. Thus, this chapter is focused on the investigation of existing FL‐based Intrusion Detection Systems (FL‐based IDSs) in order to provide a roadmap to support the design and development of effective, efficient, and privacy‐preserving IDSs for protecting the emerging disruptive massive IoT applications in the 6G era.